[PATCH][ARM][1/2] Implement lceil, lfloor, lround optabs with new ARMv8-A instructions

[PATCH][ARM][1/2] Implement lceil, lfloor, lround optabs with new ARMv8-A instructions

[Kyrill Tkachov]

[-- Attachment #1: Type: text/plain, Size: 833 bytes --]

Hi all,

This patch implements the {lceil, lfloor, lround}si{sf, df}2 optabs in a 
similar way to fcvt in aarch64. We use the new ARMv8 FP convert with 
rounding instructions vcvt{a,p,m} for that.

Bootstrapped and tested on arm-none-linux-gnueabihf.

Ok for trunk?

Thanks,
Kyrill

2014-09-02  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>

     PR target/62275
     * config/arm/iterators.md (FIXUORS): New code iterator.
     (VCVT): New int iterator.
     (su_optab): New code attribute.
     (su): Likewise.
     * config/arm/vfp.md (l<vrint_pattern><su_optab><mode>si2): New pattern.

2014-09-02  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>

     PR target/62275
     * gcc.target/arm/lceil-vcvt_1.c: New test.
     * gcc.target/arm/lfloor-vcvt_1.c: Likewise.
     * gcc.target/arm/lround-vcvt_1.c: Likewise.

[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #2: arm-vcvt.patch --]
[-- Type: text/x-patch; name=arm-vcvt.patch, Size: 4990 bytes --]

commit ba3ec05be54d74ee53d287dfa4eb7b5508292e59
Author: Kyrylo Tkachov <kyrylo.tkachov@arm.com>
Date:   Thu Aug 21 12:28:00 2014 +0100

    [ARM] Implement ARMv8-A vcvt* optabs

diff --git a/gcc/config/arm/iterators.md b/gcc/config/arm/iterators.md
index 6fe6eef..f7e0e14 100644
--- a/gcc/config/arm/iterators.md
+++ b/gcc/config/arm/iterators.md
@@ -194,6 +194,9 @@ (define_code_iterator SE [sign_extend zero_extend])
 ;; Right shifts
 (define_code_iterator rshifts [ashiftrt lshiftrt])
 
+;; Iterator for integer conversions
+(define_code_iterator FIXUORS [fix unsigned_fix])
+
 ;; Binary operators whose second operand can be shifted.
 (define_code_iterator shiftable_ops [plus minus ior xor and])
 
@@ -215,6 +218,8 @@ (define_code_attr arith_shift_insn
 (define_int_iterator VRINT [UNSPEC_VRINTZ UNSPEC_VRINTP UNSPEC_VRINTM
                             UNSPEC_VRINTR UNSPEC_VRINTX UNSPEC_VRINTA])
 
+(define_int_iterator VCVT [UNSPEC_VRINTP UNSPEC_VRINTM UNSPEC_VRINTA])
+
 (define_int_iterator NEON_VRINT [UNSPEC_NVRINTP UNSPEC_NVRINTZ UNSPEC_NVRINTM
                               UNSPEC_NVRINTX UNSPEC_NVRINTA UNSPEC_NVRINTN])
 
@@ -519,6 +524,13 @@ (define_code_attr optab [(ltu "ltu") (geu "geu")])
 ;; Assembler mnemonics for signedness of widening operations.
 (define_code_attr US [(sign_extend "s") (zero_extend "u")])
 
+;; Signedness suffix for float->fixed conversions.  Empty for signed
+;; conversion.
+(define_code_attr su_optab [(fix "") (unsigned_fix "u")])
+
+;; Sign prefix to use in instruction type suffixes, i.e. s32, u32.
+(define_code_attr su [(fix "s") (unsigned_fix "u")])
+
 ;; Right shifts
 (define_code_attr shift [(ashiftrt "ashr") (lshiftrt "lshr")])
 (define_code_attr shifttype [(ashiftrt "signed") (lshiftrt "unsigned")])
diff --git a/gcc/config/arm/vfp.md b/gcc/config/arm/vfp.md
index 90e001c..a203449 100644
--- a/gcc/config/arm/vfp.md
+++ b/gcc/config/arm/vfp.md
@@ -1306,6 +1306,18 @@ (define_insn "<vrint_pattern><SDF:mode>2"
    (set_attr "conds" "<vrint_conds>")]
 )
 
+;; Implements the lround, lfloor and lceil optabs.
+(define_insn "l<vrint_pattern><su_optab><mode>si2"
+  [(set (match_operand:SI 0 "register_operand" "=t")
+        (FIXUORS:SI (unspec:SDF
+                        [(match_operand:SDF 1
+                           "register_operand" "<F_constraint>")] VCVT)))]
+  "TARGET_HARD_FLOAT && TARGET_FPU_ARMV8 <vfp_double_cond>"
+  "vcvt<vrint_variant>%?.<su>32.<V_if_elem>\\t%0, %<V_reg>1"
+  [(set_attr "predicable" "no")
+   (set_attr "type" "f_cvtf2i")]
+)
+
 ;; MIN_EXPR and MAX_EXPR eventually map to 'smin' and 'smax' in RTL.
 ;; The 'smax' and 'smin' RTL standard pattern names do not specify which
 ;; operand will be returned when both operands are zero (i.e. they may not
diff --git a/gcc/testsuite/gcc.target/arm/lceil-vcvt_1.c b/gcc/testsuite/gcc.target/arm/lceil-vcvt_1.c
new file mode 100644
index 0000000..bbe4271
--- /dev/null
+++ b/gcc/testsuite/gcc.target/arm/lceil-vcvt_1.c
@@ -0,0 +1,21 @@
+/* { dg-do compile } */
+/* { dg-require-effective-target arm_v8_vfp_ok } */
+/* { dg-options "-O2 -march=armv8-a" } */
+/* { dg-add-options arm_v8_vfp } */
+
+int
+foofloat (float x)
+{
+  return __builtin_lceilf (x);
+}
+
+/* { dg-final { scan-assembler-times "vcvtp.s32.f32\ts\[0-9\]+, s\[0-9\]+" 1 } } */
+
+
+int
+foodouble (double x)
+{
+  return __builtin_lceil (x);
+}
+
+/* { dg-final { scan-assembler-times "vcvtp.s32.f64\ts\[0-9\]+, d\[0-9\]+" 1 } } */
diff --git a/gcc/testsuite/gcc.target/arm/lfloor-vcvt_1.c b/gcc/testsuite/gcc.target/arm/lfloor-vcvt_1.c
new file mode 100644
index 0000000..88671d3
--- /dev/null
+++ b/gcc/testsuite/gcc.target/arm/lfloor-vcvt_1.c
@@ -0,0 +1,21 @@
+/* { dg-do compile } */
+/* { dg-require-effective-target arm_v8_vfp_ok } */
+/* { dg-options "-O2 -march=armv8-a" } */
+/* { dg-add-options arm_v8_vfp } */
+
+int
+foofloat (float x)
+{
+  return __builtin_lfloorf (x);
+}
+
+/* { dg-final { scan-assembler-times "vcvtm.s32.f32\ts\[0-9\]+, s\[0-9\]+" 1 } } */
+
+
+int
+foodouble (double x)
+{
+  return __builtin_lfloor (x);
+}
+
+/* { dg-final { scan-assembler-times "vcvtm.s32.f64\ts\[0-9\]+, d\[0-9\]+" 1 } } */
diff --git a/gcc/testsuite/gcc.target/arm/lround-vcvt_1.c b/gcc/testsuite/gcc.target/arm/lround-vcvt_1.c
new file mode 100644
index 0000000..8b1f6a7
--- /dev/null
+++ b/gcc/testsuite/gcc.target/arm/lround-vcvt_1.c
@@ -0,0 +1,21 @@
+/* { dg-do compile } */
+/* { dg-require-effective-target arm_v8_vfp_ok } */
+/* { dg-options "-O2 -march=armv8-a -ffast-math" } */
+/* { dg-add-options arm_v8_vfp } */
+
+int
+foofloat (float x)
+{
+  return __builtin_lroundf (x);
+}
+
+/* { dg-final { scan-assembler-times "vcvta.s32.f32\ts\[0-9\]+, s\[0-9\]+" 1 } } */
+
+
+int
+foodouble (double x)
+{
+  return __builtin_lround (x);
+}
+
+/* { dg-final { scan-assembler-times "vcvta.s32.f64\ts\[0-9\]+, d\[0-9\]+" 1 } } */

Re: [PATCH][ARM][1/2] Implement lceil, lfloor, lround optabs with new ARMv8-A instructions

[Ramana Radhakrishnan]

On 02/09/14 16:34, Kyrill Tkachov wrote:
> Hi all,
>
> This patch implements the {lceil, lfloor, lround}si{sf, df}2 optabs in a
> similar way to fcvt in aarch64. We use the new ARMv8 FP convert with
> rounding instructions vcvt{a,p,m} for that.
>
> Bootstrapped and tested on arm-none-linux-gnueabihf.
>
> Ok for trunk?

Ok .

Ramana

>
> Thanks,
> Kyrill
>
> 2014-09-02  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
>
>       PR target/62275
>       * config/arm/iterators.md (FIXUORS): New code iterator.
>       (VCVT): New int iterator.
>       (su_optab): New code attribute.
>       (su): Likewise.
>       * config/arm/vfp.md (l<vrint_pattern><su_optab><mode>si2): New pattern.
>
> 2014-09-02  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
>
>       PR target/62275
>       * gcc.target/arm/lceil-vcvt_1.c: New test.
>       * gcc.target/arm/lfloor-vcvt_1.c: Likewise.
>       * gcc.target/arm/lround-vcvt_1.c: Likewise.
>